# `LangChain.OpenTelemetry.ProviderMapping`
[🔗](https://github.com/brainlid/langchain/blob/v0.9.3/lib/open_telemetry/provider_mapping.ex#L1)

Maps LangChain provider strings to OpenTelemetry GenAI Semantic Convention
`gen_ai.provider.name` values.

Values are verified against the well-known `gen_ai.provider.name` set in the
GenAI semantic conventions. Providers with no well-known value (`ollama`,
`orq`, `bumblebee`) intentionally pass through unchanged — the spec permits
custom identifiers for providers it doesn't enumerate.

## `google` → `gcp.gemini` (reconciliation note)

The spec defines two Google values: `gcp.gemini` ("Gemini", i.e. the Gemini
API / Google AI Studio served from `generativelanguage.googleapis.com`) and
`gcp.gen_ai` ("any Google generative AI endpoint", a generic fallback).
`LangChain.ChatModels.ChatGoogleAI` targets the Gemini API specifically, so we
deliberately emit the more precise `gcp.gemini` rather than the generic
`gcp.gen_ai`. (The sibling `req_llm` library maps its `:google` provider to the
generic `gcp.gen_ai`; the divergence is intentional — LangChain's value is the
more specific, spec-correct one for the Gemini endpoint.) Vertex AI is a
separate provider (`ChatVertexAI` → `gcp.vertex_ai`).

See: https://opentelemetry.io/docs/specs/semconv/gen-ai/

# `to_otel`
[🔗](https://github.com/brainlid/langchain/blob/v0.9.3/lib/open_telemetry/provider_mapping.ex#L59)

```elixir
@spec to_otel(String.t()) :: String.t()
```

Returns the OTel `gen_ai.provider.name` for a LangChain provider string.

Unknown providers are passed through unchanged.

## Examples

    iex> LangChain.OpenTelemetry.ProviderMapping.to_otel("openai")
    "openai"

    iex> LangChain.OpenTelemetry.ProviderMapping.to_otel("google")
    "gcp.gemini"

    iex> LangChain.OpenTelemetry.ProviderMapping.to_otel("custom_provider")
    "custom_provider"

---

*Consult [api-reference.md](api-reference.md) for complete listing*